AI debt issuance surges as hyperscaler cash flows lag

AI capital expenditure is outgrowing the free cash flows of hyperscalers, meaning these companies can no longer fund expansion solely from existing businesses. Between half a trillion and one trillion dollars of debt has been issued around AI this year, alongside well over a trillion dollars in leases and future purchase commitments. Anthropic has made non-cancellable commitments for hundreds of billions to over a trillion dollars in future capacity. This financial structure makes capital market sentiment critical, a significant change from 2024 when equity growth was the primary focus. Credit investors now prioritize cash flow coverage; for Google, ad business cash flows support AI debt even if AI initiatives fail, but neoclouds and Oracle lack such a fallback, creating correlated downside risk. NVIDIA’s argument that chips are fungible collateral has been rejected by analysts, who note that if a neocloud fails due to lack of demand, no other buyer will emerge for the hardware.

Recent market reactions underscore this vulnerability. Oracle announced plans to raise $20 billion through new equity and debt, causing its stock to fall 8.8%. Anthropic filed an updated prospectus revealing payment obligations of approximately $895 billion, almost entirely for computing power, against a quarterly revenue of $17 billion. The company’s year-over-year growth slowed from 144% to 47%. In related moves, Marvell agreed to acquire Lumentum for $115 billion in stock to secure laser components for AI data centers. Marvell’s stock fell 12.7% and Lumentum’s fell 18.6% due to investor frustration with circularity in the AI sector. The Philadelphia Semiconductor Index lost more than 9% in five days, despite WTI crude falling 7% to $96.50 per barrel.